Year: 2025
Type: conference-paper
Abstract: Assessing children’s pronunciation presents a different set of challenges than assessing adults’ language. Patterns of errors in children’s speech are highly variable, and sources of the errors may be because of developmental stage, speech sound disorders, and language learning status. In school settings, different professionals work with children depending on their needs, including Speech-Language Pathology (SLP) professionals and English as a Second Language (ESL) teachers. Computers, on the other hand, can be trained to identify a wide array of patterns quite quickly. As part of a larger project to create an automated pronunciation diagnostic test for children, we created a test for children 5-7 years old and compared automatic evaluations of 26 children’s pronunciation to human evaluations. While the human evaluations showed high agreement, the automatic mispronunciation detection system was inconsistent in its evaluations, making pattern detection for pronunciation screening impossible. We suggest the development of specialized pronunciation detection systems.
DOI: 10.31274/psllt.18696
Authors: Mackenzie Novotny, Carolyn Richie, John M Levis, Evgeny Chukharev
Institution: Iowa State University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411367900